Predictive lithological mapping based on geostatistical joint modeling of lithology and geochemical element concentrations

Abstract

The spatial analysis and interpretation of lithological and geochemical sampling information are central in mineral prospecting and initial geological-mining exploration to delineate exploration targets and locate economic mineralization. This work compares two geostatistical approaches for the spatial prediction of lithological classes through a case study in mineral prospection, considering lithological and geochemical information at a set of surface samples. Both approaches calculate the probabilities of occurrence of the lithological classes at unsampled locations and select the most probable class as the predicted lithology. A split-sample technique is used to assess their performance, with the predictions being made at a testing data subset on the basis of the information of a training subset. The first approach relies on a cokriging of the lithological class indicators and yields an accuracy score (percentage of matches between true and predicted lithological classes) of 90.5%, while the second approach, consisting of a plurigaussian modeling of the classes, increases this score to 92.6%. Unlike the former approach, it also provides consistent outcomes of both the lithological classes and the geochemical covariates, which is valuable for mineral prospectivity mapping.